DYNAMIC CONVOLUTIONAL NEURAL NETWORKS FOR SENTIMENT ANALYSIS

Neural networks are a subset of artificial intelligence that use a simplified mathematical model of the human brain in order to build models used for solving many problems, such as recognizing an image, grouping data into categories, or examining text. Many types of neural networks have been used to analyze text for tasks such as question answering, machine translation, summarization, or classification. The convolutional neural network (CNN) has been thought to perform poorly on text-based tasks. However, the dynamic convolutional neural network (DCNN), a modification to the traditional CNN, has previously shown promise in text-related tasks. Thus, the effectiveness of the DCNN for sentiment analysis, or deciding if a short body of text expresses positive or negative sentiment, was examined and compared to the CNN and the support vector machine (SVM), which are two other common architectures used for this application. Using TensorFlow™ and Python 3™, a DCNN, CNN, and SVM were implemented for sentiment analysis. The three architectures were trained using approximately 11,000 previously harvested short movie reviews from Rotten Tomatoes® and tested using approximately 2,000 previously harvested short movie reviews from IMDB®. After training, the DCNN performed significantly better than both the CNN and the SVM, with accuracies of 85% vs. 71.9% and 72%, respectively (N-1 chi squared test, DCNN compared to CNN, p